Same merits, same chances.

For recruitment and staffing agencies and employers that use AI to rank or filter CVs. We test whether your tool treats an applicant differently because of age, so you can show clients and applicants that screening is fair, or fix it if it isn’t.

Ask us a question

Test 1, in progress

Does AI rank a 55 year old lower than an identical 35 year old? Our method is fixed and registered before we look at any data, so we cannot adjust it to the outcome. That is why no results are shown yet. The results will be published under Tests, including if we find no difference.

Read the method

Age 35

Age 55

Identical in every way except one.

Illustration. No results shown.

What we do

We start with AI tools used in recruitment. We test them with pairs of fictional applicants who are identical in every way except one, such as age, and measure whether the outcome changes.

See audit levels

Why test

  • The responsibility is yours. An employer may not discriminate against job applicants (Swedish Discrimination Act, chapter 2, section 1). The prohibition covers both direct and indirect discrimination, with the exceptions set out in the Act. The employer remains responsible when an external tool is used in screening. Recruitment agencies that place candidates are also covered by a prohibition against discrimination (chapter 2, section 9).
  • Recruitment is part of active measures. Employers must investigate whether there are risks of discrimination, and that work must cover recruitment (chapter 3, sections 2 and 5). A test shows how the tool treated applicants in the cases tested and gives you concrete questions to put to your vendor. It does not replace your own work on active measures.
  • You could miss qualified candidates. If age affects the ranking, applicants with the right qualifications may be ranked lower and overlooked in screening.
  • You can share the results with your clients. Recruitment and staffing agencies can choose an option allowing them to share the full report with their clients, so the client can see how the screening tool was tested. The option is quoted separately, and when shared, the report must be shown in full.

The legal provisions are summarized. The audit is not a legal assessment.

Open method

Our method is open. Every test we initiate ourselves is published with its limitations, and a null result is published as a null result. Audits ordered by clients are confidential and published only with the client's written consent.

Why independent review

  • Independent. A test that a vendor runs on its own tool is hard for buyers and applicants to trust. We do not develop, sell or use any of the tools we test.
  • Method that holds. Paired testing requires that the pairs differ in only one characteristic, that the number of pairs is set in advance so that a difference of a given size would very likely show, and that the variation due to chance is measured separately. Our rule: no test we publish is released until an independent statistician and an independent subject matter expert have reviewed it.
  • Checkable. The statistics are computed in code saved with the report. Running the same code on the same collected responses always gives the same result.

Contact

Organizations that want to know how their own tools and processes treat people are welcome to get in touch.

hello@angoff.se